Syntax reading group, 4/11 – Bjorkman

abstract: In Minimalist frameworks, the operation Agree establishes relationships between defective probes (unvalued or uninterpretable features) and non-defective goals (valued or interpretable). While the original definition of Agree required probes to search downward for goals (Chomsky 1995 et seq.), more recent work has suggested instead that probes search upwards(Wurmbrand 2012, Zeijlstra 2012, a.o.), or that the direction of Agree is variable (Baker 2008, Merchant 2011). These proposals have often drawn on different empirical domains (e.g. φ-agreement vs. verbal inflection or negative concord), raising the question of whether a single model of Agree can account for all syntactic feature relationships.

The most serious empirical issue for “Upwards Agree” models is the existence of long-distance agreement (LDA) patterns, cases where there is no point in the derivation where the goal DP c-commands the probing head. Such cases appear to be incompatible with Upwards Agree theories, as observed by Preminger (2014). This talk argues, however, that a slightly modified version of Upwards Agree not only can account for LDA, but may in fact provide a better account of the typology of LDA patterns. The main modification is to distinguish feature checking (accomplished by Agree) from feature valuation, but to restrict valuation to instances where two heads stand in some Agree (i.e. checking) relation, though potentially for a feature other than the one being valued. This predicts that LDA should be possible only when a DP stands in an independent Agree relationship with the agreeing head, whether for Case or information structural features. We illustrate how this version of Upwards Agree accounts for at least three subtypes of LDA: nominative object agreement in Icelandic, agreement into non-finite clauses in Hindi, and agreement into finite embedded clauses in Tsez.